On Mon, May 10, 2021 at 11:00:59AM -0700, Martin KaFai Lau wrote:
> On Sat, May 08, 2021 at 12:53:01PM +0200, Michal Suchánek wrote:
> > On Fri, May 07, 2021 at 10:03:21PM -0700, Martin KaFai Lau wrote:
> > > On Thu, May 06, 2021 at 07:54:37PM -0700, Nathan Chancellor wrote:
> > > > On Thu, May 06, 2021 at 01:56:22PM -0700, Martin KaFai Lau wrote:
> > > > > BTF is currently generated for functions that are in ftrace list
> > > > > or extern.
> > > > >
> > > > > A recent use case also needs BTF generated for functions included in
> > > > > allowlist. In particular, the kernel
> > > > > commit e78aea8b2170 ("bpf: tcp: Put some tcp cong functions in allowlist for bpf-tcp-cc")
> > > > > allows bpf program to directly call a few tcp cc kernel functions. Those
> > > > > kernel functions are currently allowed only if CONFIG_DYNAMIC_FTRACE
> > > > > is set to ensure they are in the ftrace list but this kconfig dependency
> > > > > is unnecessary.
> > > > >
> > > > > Those kernel functions are specified under an ELF section .BTF_ids.
> > > > > There was an earlier attempt [0] to add another filter for the functions in
> > > > > the .BTF_ids section. That discussion concluded that the ftrace filter
> > > > > should be removed instead.
> > > > >
> > > > > This patch is to remove the ftrace filter and its related functions.
> > > > >
> > > > > Number of BTF FUNC with and without is_ftrace_func():
> > > > > My kconfig in x86: 40643 vs 46225
> > > > > Jiri reported on arm: 25022 vs 55812
> > > > >
> > > > > [0]: https://lore.kernel.org/dwarves/20210423213728.3538141-1-kafai@xxxxxx/
> > > > >
> > > > > Cc: Andrii Nakryiko <andrii@xxxxxxxxxx>
> > > > > Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
> > > > > Signed-off-by: Martin KaFai Lau <kafai@xxxxxx>
> > > >
> > > > This fixes an issue with Fedora's s390x config that CKI noticed:
> > > >
> > > > https://groups.google.com/g/clang-built-linux/c/IzthpckBJvc/m/MPWGDmXiAwAJ
> > > >
> > > > Tested-by: Nathan Chancellor <nathan@xxxxxxxxxx> # build
> > > Thanks all for reviewing and testing.
> > >
> > > In my cross compile ppc64 test, it does not solve the issue.
> > > The problem is the tcp-cc functions (e.g. "cublictcp_*")
> > > are not STT_FUNC in ppc64, so they are not collected in collect_function().
> > > The ".cubictcp_*" is STT_FUNC though.
> > >
> > > Since only the x86 (64 and 32) bpf jit can call these tcp-cc functions now
> > > and there is no usage for adding them to .BTF_ids for other ARCHs,
> > > I have post a patch to limit them to x86:
> > > https://lore.kernel.org/bpf/20210508005011.3863757-1-kafai@xxxxxx/
> >
> > That's probably not the way to go. If function symbols start with a dot
> > in ppc64 elfv1 abi then pahole should learn to add a dot for ppc64 elfv1
> > abi.
> >
> > Or we can build ppc64 BE using the elfv2 abi and depend on elfv2 abi for
> > BTF on ppc64. The patch for enabling elfv2 for BE is currently under
> > discussion and I have the patch that adds the dependency ready as well.
> ic. I just learned the elfv2/LE vs elfv1/BE difference in ppc64.
> elfv2 BE ppc64 support is on its way, so it is better to have DEBUG_INFO_BTF
> depending on the ppc64-elfv2 related kconfig(s) instead of making an exception
> handling in pahole.
>
> If you have the BTF ppc64-elfv2 dependency patch ready, does it make
> sense to land this patch first independent of the ppc64 elfv2/BE
> effort?
It kind of depends on the result of the config option name bikeshedding
so it has to be merged after.
Thanks
Michal
